An integrated method to detect the incipient degradation of bearings by vibration analysis and feature extraction

As fundamental mechanical components, bearings are vital to various rotating machinery. By monitoring their conditions and predicting their remaining useful lives (RULs), some proactive maintenance may be done to reduce the unscheduled machine downtime and even catastrophes. One of the key issues in bearing prognostics is to detect the defect at the early stage so as to alert the operator, maintenance personnel and logistics personnel before the defect develops into an unrepairable failure. Since the signature of an incipient defect is relatively weak and masked by strong noise, robust signal de-noising and feature extraction methods are crucial to bearing prognostics. This paper introduces an integrated method to detect the incipient defects of bearings by vibration analysis and feature extraction. First, the vibration signals collected from a bearing accelerated life test are de-noised by a 3-layer wavelet packet decomposition (WPD). Second, an empirical mode decomposition (EMD) is performed to decompose the de-noised signal into intrinsic mode functions (IMFs) so as to extract features of bearing faults, and then some rules are defined to select the IMFs containing the fault information. Third, energy moments of the selected IMFs are extracted as features to detect the incipient defects. Fourth, the advantage of this energy moment is demonstrated by a comparative analysis with the classic time domain statistical features. In the end, the summary of this work and some future aspects are given.